{"id":"W1975396736","doi":"10.1109/mdt.2010.73","title":"Time to retire our benchmarks","year":2010,"lang":"en","type":"article","venue":"IEEE Design & Test of Computers","topic":"Distributed and Parallel Computing Systems","field":"Computer Science","cited_by":2,"is_retracted":false,"has_abstract":true,"ca_institutions":"","funders":"","keywords":"Argument (complex analysis); Computer science; Quarter (Canadian coin); Sentence; Artificial intelligence; History","routes":{"ca_aff":false,"ca_fund":false,"ca_venue":false,"about_ca":true,"invisible_to_affiliation_only":true},"retraction":null,"screen":null,"direct_labels":[],"prediction":{"model_version":"metacan-v3-hybrid-931329e0061c","candidate_categories":["metaresearch"],"consensus_categories":[],"category_scores_codex":[0.00977832,0.001788388,0.001366815,0.003852775,0.002450041,0.00651602,0.003567571,0.001602122,0.1356947],"category_scores_gemma":[0.04232169,0.0007975542,0.001268727,0.00284465,0.000744142,0.006335024,0.004303062,0.00543086,0.1152336],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.003153126,"about_ca_system_score_gemma":0.004315678,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.007597779,"about_ca_topic_score_gemma":0.01093874,"domain_scores_codex":[0.9909685,0.00121252,0.000393594,0.0004940469,0.005367212,0.001563999],"domain_scores_gemma":[0.9680797,0.002447705,0.0006566011,0.003852018,0.0188982,0.006065736],"domain_codex":null,"domain_gemma":"evaluation","domain_candidate":"evaluation","domain_consensus":null,"study_design_codex":"not_applicable","study_design_gemma":"theoretical_or_conceptual","study_design_scores_codex":[0.0001295266,0.00008815537,0.0003992796,0.0001107197,0.00001147757,0.00007748704,0.00005962251,0.000507414,0.001251025,0.00579806,0.8776432,0.1139241],"study_design_scores_gemma":[0.00006481512,0.0002137128,0.001241449,0.0001598589,0.00001858919,0.0001294262,0.0001704749,0.0009031508,0.001832259,0.007262269,0.987948,0.00005604174],"study_design_candidate":"theoretical_or_conceptual","study_design_consensus":null,"genre_codex":"other","genre_gemma":"commentary","genre_scores_codex":[0.02955712,0.02473416,0.07945557,0.1468921,0.1381817,0.00181755,0.02013375,0.04795375,0.5112743],"genre_scores_gemma":[0.1213768,0.0136874,0.09267772,0.07045113,0.02838227,0.00190694,0.06806133,0.03036121,0.5730951],"genre_candidate":"commentary","genre_consensus":null,"teacher_disagreement_score":0.9902217,"threshold_uncertainty_score":0.4539441,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.01433978754296901,"score_gpt":0.2362975613649401,"score_spread":0.2219577738219711,"validation_status":"score_only:v0-immature-baseline","note":"Baseline scores from an immature model (maturity gate not passed). Scores rank; they never assert a category."}}